Urgent.News

What's breaking now, across thousands of outlets.

Tech

Agent Access Without Shared Passwords: A Scoped Grant Pattern for Banks and Fintech

On October 22, Reuters and Axios reported that major US banks are restricting outside AI agents from customer accounts. As reported, customers share a username and password with an AI tool, the agent logs in, and the bank sees only the customer. Some agents tried actions the customer had not asked for. Most banks are moving towards read-only access for agents as a middle ground. Read-only is a…

Reuters and Axios reported on October 22 that major US banks are limiting access for outside AI agents to customer accounts. Customers currently share usernames and passwords with AI tools, which log in and banks only see the customer. Some agents performed actions the customer did not request. Most banks are now aiming for agents to have read-only access as a compromise.

However, the fundamental issue is that banks cannot differentiate agents from users. This article outlines a simple pattern to address this, using code I tested prior to drafting the piece. It is a learning example, not a finished product.

The problem with using shared logins lies in the fact that a password confirms an account holder's existence but provides no information about who is acting (person or software), on whose behalf, for what purpose, within what limits, or for how long. Once the agent uses the customer's login, all relevant details merge into a single name. The log records "customer logged in" without additional context. To resolve this, a four-rule pattern is proposed:

1. The agent possesses a distinct identity separate from the customer.

2. The customer issues a grant specifying the agent, actions, limits, expiry, and responsible company.

3. Any transaction requiring funds necessitates approval from the customer's app using a separate key.

4. Each log line identifies two actors: the acting agent and the customer it represents.

Step 1 involves creating helper functions and a signing method. The code uses base64 encoding, HMAC signing, JSON serialization, and timestamps. Bank and customer application keys are separate, held in a KMS or HSM in production. Two functions are defined: one to create a sealed grant and another to verify sealed approvals.

Step 2 explains how to issue a grant and obtain customer approval. A grant defines the scope, limits, and expiration date, signed by the bank. Approval involves the customer's app, signed using a distinct key, and never entered into the agent's prompt.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in Tech

IoT for Beginners: From MQTT to Device Management, Platforms, and Edge Computing

IoT Basics · MQTT · Edge Computing | FastBee Technical Team | Published October 26, 2026 · Reviewed October 27, 2026 The Internet of Things (IoT) is a system of connected devices, networks, and…

  • MQTT moves publish/subscribe messages between devices and applications
  • IoT platform organizes device identity, data models, rules, and applications
  • Edge computing needed for local responses, intermittent connectivity, or on-site processing

git-ofrenda: an altar for the code your repo buried, with epitaphs by Gemma on a plain CPU 🕯️

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine What I Built git-ofrenda 🕯️🌼 builds an ofrenda for the code your repo buried.

  • git-ofrenda is a Python tool for creating an ofrenda for deleted files in Git repositories
  • Open-weight language model Gemma 3 writes epitaphs and remembrance for each file
  • Interactive séance allows users to ask questions based on file's Git history

How Browser-Based Reaction Time Tests Work

A reaction time test looks simple: wait for a color change, then click. Behind the result are a timer, browser rendering, input events, and measurement limits.

  • Browser-based reaction tests monitor color shift and button press.
  • JavaScript code initiates trials with random delay and captures timestamps.
  • Randomized delay prevents prediction of signal appearance.

High Roughness, Sharp Highlights: Auditing a glTF Material in Three.js

An imported car-body model still showed sharp highlights on its hood and rear fender after switching to a preset labeled Matte . The overall reflection became weaker, but some bright patches remained.

  • BMW M4 G82 ADRO-kit car body imported into Three.js environment
  • Matte preset showed sharp highlights on hood and rear fender
  • Clearcoat extension explained remaining highlights in rendered image

More from Tuesday 27 October →